JPH02156331A - Method and device for aiding judgement - Google Patents

Abstract

PURPOSE:To obtain a device which prevents expert's premature judgement, oversight, or the like and is a capable helper and is free from unnecessary operation by verifying expert's judgement and spontaneously presenting another answer or a competitive answer. CONSTITUTION:A certification processing 305 is started with a character string indicating a potential answer as the argument by a control processing 303. When this answer is found in a work memory 309, certification results in suc cess, but when it cannot be found there, a rule having this answer as a conclu sion character string is extracted from a rule group 308, and success or failure of a condition character string and a condition formula of this rule is verified. A leading-out means 306 compares the rule, whose all condition character strings and condition formulas are satisfied, with an answer candidate group 310 to discriminate the rule while referring to character strings and variable values stored in the work memory 309, and the rule is transferred to the control processing 303 when it is an answer, but the rule is stored in the memory 309 as an intermediate result when it is not an answer. Thus, the device which is a capable helper and is free from unnecessary operation is obtained.

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] The present invention relates to a data processing system, such as a reasoning machine or an expert system, that makes decisions regarding problems based on a specific set of rules, and in particular, to Rather, it concerns a system of this kind that is suitable for use as a specialist's assistant.

[Conventional technology]

First, an outline of the inference machine will be explained. Reasoning machines receive and store knowledge in the field of interest in the form of rules. When a user inputs the problem he/she wants to solve in the form of data into an inference machine, the inference machine performs the specified inference using the aforementioned rule group. Conventional inference machines are described in Information Processing Society of Japan, Vol. 26, No. 12 (1985).
As discussed in J.D., pp. 1487-1496, two types of inferences can be made:

(1) Input the proof processing question and the user's answer to the question, and determine whether the user's answer is correct or incorrect. Specifically, data representing the problem and the user's answer are input, and it is verified whether the latter can be derived from the former through a set of rules without contradiction. Hereinafter, the user's answer to the question will be referred to as the user's "expected answer."

The proof process is a process of verifying the predicted solution.

(2) Derivation processing Enter a question and derive a new answer.

Specifically, data representing a problem is input, and an answer is derived from this data via a set of rules.

For example, in a medical diagnostic system, body temperature is 37.5
℃, pale complexion, and other symptoms are used as problem data, and the name of a disease called a cold is used as a predicted answer.The proof process involves verifying whether the user's predicted answer is correct using the data and a set of rules. On the other hand, from these data and rules,
The derivation process is to newly estimate the name of a disease such as a cold or influenza.

Efforts have been underway since the 1970s to have reasoning machines perform the intellectual tasks performed by experts. However, as a result of many applied experiments, it has become clear that it is difficult to develop reasoning machines that can completely perform the intellectual tasks of experts, except in some fields that are suitable for computerization.

Therefore, in recent years, consideration has been given to using inference machines as assistants rather than substitutes for experts.

When an inference machine is used as an assistant to an expert, different functions are required than when it is used as a substitute for the expert. As an assistant, a reasoning machine needs the ability to supplement the judgment of experts, rather than the ability to draw conclusions on its own. In other words, it is necessary to have a function that not only verifies the judgment of experts but also provides alternative or counter-solutions as advice in order to prevent premature mistakes or oversights and lead to the most appropriate conclusion.

Such functionality can be achieved in part simply by using individual proof or derivation processes. For example, according to the proof process, it is possible to verify the expert's judgment by inputting the situation as data and the expert's judgment regarding the data as a predicted solution. Moreover, according to the derivation process, the situation can be input as data and several possible judgments can be output.

[Problem to be solved by the invention]

However, the conventional technology only provides individual partial functions, and cannot integrate these functions to provide sufficient functions as an assistant system. That is,
Instead of the necessary functions being activated automatically as appropriate, the expert had to select the necessary functions himself and activate them individually. Therefore, when actually using it as an assistant, there were the following inconveniences.

First, the ability to point out premature mistakes or oversights by experts is weak.
After verifying the expert's judgment and determining that it is correct, a competent assistant should not only express agreement with the expert's judgment, but also explore other possible judgments (alternative solutions), if any. , this should be shown.

For example, a doctor determines that a patient has a ``cold'' based on his symptoms.

When an assistant system is requested to perform this verification, the assistant system should look for the possibility of other diseases even if no error is detected in the doctor's judgment. However, in the conventional technology, when the expert verifies the judgment using the proof process and receives a positive answer, the system will not work unless the expert is cautious and starts a new derivation process in search of another possibility. Does nothing, therefore cannot automatically prevent oversight by experts. For example, if a patient's symptoms indicate that ``tuberculosis'' is more likely than a ``cold,'' the system will automatically point this out. There's nothing to do.

In addition, there is no function to proactively advise against errors in the judgment of experts.If an expert's judgment is verified and determined to be incorrect, a competent assistant should not only express opposition, but also The judgment (counter solution) should be shown. However, in the prior art, the system simply expresses the opposition;
Such a system, which does not show its own judgment and requires the expert to restart the derivation process in order to obtain a counter solution, is extremely unfriendly as an assistant.

As described above, the conventional technology cannot provide sufficient functions to be used as an assistant to an expert, and requires careful attention and redundant operations on the part of the expert to be useful as an assistant. Ta.

The purpose of the present invention is to verify the expert's judgment and spontaneously present an alternative or counter solution, thereby preventing the expert from making premature decisions or overlooking the matter, or providing specific suggestions for the correct answer. and is extremely capable as an assistant, and
The objective is to provide a decision support system that is efficient in operation.

[Means to solve the problem]

The present invention performs proof processing in response to a startup command, and starts derivation processing in conjunction with the proof processing. The proof process attempts to prove the input predicted solution using the input data, the derivation process uses this same input data to derive all possible solutions, and at least one process attempts to prove the input predicted solution. Where necessary, the processing uses additional data requested and obtained from the user during the process. The solution obtained by the derivation process is output in association with the result of the proof process.

For example, if the proof process succeeds in proving the predicted solution,
Solutions different from the predicted solution are output as separate solutions, and if the proof process fails to prove the predicted solution, all solutions are output as counter solutions.

A plurality of predicted solutions may be input and all of them may be verified in response to a single activation. Another variation is
Instead of directly outputting alternative solutions or counter solutions, information indicating the presence or absence of alternative solutions or counter solutions may be output, and these solutions may be output when requested by the user.

Either the proof process or the derivation process may be performed first, but
It is advantageous to carry out the proof process before the derivation process. Alternatively, the derivation process may be performed in parallel with the proof process. In the former case, the derivation process utilizes additional data requested and entered during the proof process; in the latter case, any one process utilizes additional data requested and entered during the other process. Take advantage of. Although the derivation process and the proof process can be performed in parallel in any time-sharing manner, it is advantageous to give priority to the proof process and perform the derivation process while the proof process is in a waiting state.

The invention can be implemented by an expert system in which rules are stored in advance, or by a reasoning machine to which at least some of the rules are supplied as needed.

The device of the present invention, which is suitable for performing derivation processing and proof processing in parallel, has a first inference unit for proof processing, a second inference unit for derivation processing, and controls the parallel operations of both inference units, as well as input/output. A control unit that performs processing;
a memory shared by the first and second inference units and the control unit.

[Effect]

According to the present invention, when a user starts the system by providing data representing a problem and a predicted solution representing his/her own judgment, the system autonomously performs both proof processing and derivation processing to support the user. Along with the pros and cons of the judgment, the system outputs its own solution as, for example, an alternative solution or a counter solution, thereby preventing the user from making premature decisions or overlooking the decision, and making it possible to efficiently reach the most appropriate conclusion. Since one process can directly use additional data input during the process of the other process, duplication of inquiries to users can be avoided and efficiency is improved.

The variant that handles multiple p-conceptions verifies each of the multiple predicted solutions and outputs the system's own solution in a single activation, so if the user reaches multiple alternative decisions, However, they can be processed together. Further, in the modification that outputs information indicating the existence of an alternative solution or a counter solution, outputting the alternative solution or a counter solution is left to the user's choice, so the user does not have to worry about excessive information.

If the derivation process is performed after the completion of the proof process, the additional data input during the proof process is saved, so the derivation process saves this additional data.

It can be used immediately without asking the user.

Additionally, when both processes are performed in parallel, the additional data input in each process is saved, so either process can save the additional data input manually in the process of the other process.
It can be used immediately without asking the user.

Especially if the derivation process is performed while the proof process is in a waiting state.

Idle time, for example between requesting additional data to the user and inputting it, is effectively utilized and is therefore more efficient compared to simple periodic time-sharing processing. In these cases, the reason why it is advantageous to give priority to the proof process is that the result of the proof process will affect the handling of the solution obtained by the derivation process, and also because informing the user of the validity of the predicted solution early will reduce user irritation. This is because there are few

Further, in the apparatus of the present invention, the proof processing and the derivation processing are simultaneously performed by separate hardware.

Both processes store rules, problem data,
The results of both processes are linked by the control unit, sharing additional data, intermediate results, and other required information. Therefore, although it goes without saying that the method of the present invention can be carried out using ordinary data processing equipment, this apparatus in particular can carry out the method of the present invention extremely efficiently.

〔Example〕

First, a first example will be described. In this example, a set of rules, data representing a situation or problem, and m-predictions are input into a reasoning machine.

Verification of predictions and derivation of alternative predictions or counter solutions are performed, and expressions of approval and predictions, or expressions of opposition and counter solutions are output.

FIG. 2 shows an example of the hardware configuration for this embodiment. The input/output terminal 201 includes, for example, a keyboard and a CR.
It is a display terminal with T, which receives a set of rules, a set of solution candidates, data, P considerations, and a startup command from the outside (user) and passes it to the central processing unit 12202.
Further, data requests generated during processing and processing results are received from the central processing unit 12202 and output to the outside. The central processing unit 202 converts the set of rules, the set of solution candidates, and data from the input/output terminal 201 into intermediate words, and then stores the rules in the storage device 203 .

Upon receiving the expected time and start command from the input/output terminal 201,
The central processing Mff 1202 first reads rules, data, and solution candidates from the storage device 203 as necessary.
The predicted interval is verified, then an alternative interval or counter solution is derived, and the verification result and the alternative interval or counter solution are sent to the input/output terminal 201. The central processing unit 202 also sends requests for data that become necessary during the verification or derivation process to the input/output terminal 201.

FIG. 3 shows the functional configuration of this embodiment. block 301
is a processing group, and input/output processing 302. Control processing 303. Analysis conversion processing 304. It consists of a proof process 305 and a derivation process 306. Block 307 is a memory, which includes a rule group section 308, a working memory 309, and a solution candidate group section 310.

The input/output processing 302 receives a set of rules, a set of solution candidates, data, a prediction interval, and a start command from the outside, outputs the set of rules to an analysis conversion process 304, and performs control processing on the prediction interval and start command. 303, and the data is subjected to control processing 303, proof processing 305 or derivation processing 3
06, and a set of solution candidates is stored in the solution candidate grouping section 310. Control processing 303 stores data from input/output processing 302 in working memory 309 . When the control processing 303 receives a prediction interval and a startup command from the input/output processing 302, it activates a proof process 305 using the prediction interval as an argument, and when the proof process is finished, a derivation process 305 is started.
Start 06. When a solution is obtained from the derivation process 306,
Control processing 303 compares it with the expected interval and discards the solution if they are equal. When the derivation process is completed, the control process 303 combines the results of the proof process and the derivation process, edits an output message, and outputs it to the input/output process 302.

The analysis conversion process 304 receives a set of rules from the input/output process 302, converts it into an intermediate language, and stores it in the rule group section 308. The proof processing 305 interprets and applies the intermediated rule set in the rule group section 308 while referring to the data stored in the working memory 309 when receiving a start command from the control processing 305 with the prediction interval as an argument. This prediction interval is verified and the result is reported to the control processing 303. During this verification process, when additional data is required, the proof processing 305 issues a request to the input/output processing 302, receives new data therefrom, and stores the number in the working memory 309.

When the derivation process 306 receives the activation command from the control process 303, the derivation process 306 interprets and applies the intermediated rule set in the rule group unit 308 while referring to the data stored in the working memory 309 and intermediate results. A solution is derived and reported to control processing 303. In this derivation process, in order to determine whether a certain derived fact can be a solution, the derivation process 306 refers to the solution candidate group 310, and if the derived Toi is included in the solution candidate set, the derivation processing 306 is judged to be a solution, and if it is not included, it is judged to be an intermediate result rather than a solution. Also, the derivation process 306
When additional data is needed, it issues a request to input/output processing 302, receives new data from it, and stores it in working memory 309.

The rule grouping unit 308 receives a set of intermediate rules from the analysis conversion process 304, stores it, and supplies the rules to the proof process 305 and the derivation process 306 according to requests. The working memory 307 receives data from the control process 303 before starting the proof process 305, and also receives data from the proof process 305 in the process of proof and derivation.
and new data from the derivation process 306, and also receives intermediate results of inference from the derivation process 306 and stores them.

The solution candidate grouping unit 310 receives from the input/output processing 302 a set of facts that can be a solution in the field of the problem targeted for inference, that is, a set of solution candidates, and stores it. For example, in the field of medical diagnosis, a set of solution candidates is a set of disease names. Furthermore, although the memory 307 is not shown,
It also includes a section for storing various other necessary information for the word processing.

FIG. 4 shows the configuration of rules used in this embodiment.

Block 401 is the condition part of the rule.

Consists of 0 or more condition character strings and conditional expressions. However, at least one of the <conditional character string> and the conditional expression> is one or more. A condition string> is a character string that represents a precondition or an intermediate result of inference. A conditional expression is an equality or inequality regarding − or more variables,
Represents preconditions for inference.

Block 402 is the conclusion part of the rule and consists of 1n eyebrow-raising conclusion character string. The conclusion character string is a character string that represents an intermediate result or solution of the inference.

FIG. 5 shows an intermediate word resulting from analysis and conversion of the rule shown in FIG. 4 by the analysis conversion processing 3011, that is, a rule intermediate word, and each rule is stored in this format in the rule group section 308. The above proof processing and derivation processing are actually performed using this intermediate language.

In this embodiment, the rule intermediate words in the rule group section 308 and the variable value information in the working memory 309 are expressed in a list structure. A list can contain any number of elements ′(′
It is written by surrounding it with and ′)′. The elements of the list may themselves be lists. The list of such inputs is between ′(′ and ′)′ and further ′(″ and ′)
It is expressed by inserting ′. The depth of nesting is arbitrary. A rule intermediate word is an intermediate word a that represents an individual rule.
! I, and the intermediate word representing the rule set is a list of rule intermediate words. Ruhr county part 308 has Ruhr intermediate language squirrel!・is memorized.

A block 501 is an intermediate word of the conditional part, and is composed of a condition character string and a conditional expression intermediate word. The condition string is
This is the same as the condition character string in block 401 in FIG. The conditional expression intermediate word is.

This is the result of converting the conditional expression in block 401 in FIG. <Intermediate conditional expression Jn> will be explained in detail later using FIG. 6.

Block 502 is an intermediate word of the conclusion part, and the conclusion character string is the same as the conclusion character string in block 402 in FIG.

FIG. 6 shows the structure of the conditional expression intermediate word in block 501 of FIG. The <conditional expression intermediate word> is the intermediate word of the conditional expression in block 1101 in FIG. Block 6
01 represents the process of determining the value of each variable in the conditional expression. get(<variable name>) represents a process of obtaining a variable value corresponding to <variable name> and substituting it to <variable name>. Block 602 is a conditional expression regarding <variable name>.

This is the same as the conditional expression in block 401 in FIG.

The proof process 305 and the derivation process 306 are as follows.

In processing the conditional expression intermediate word, first block 601
, and assigns a variable value to each <variable name>, and then processes block 602 to calculate the conditional expressions for these variable names (<variable name 1> variable name 2). ...<Variable name 1>) is established or not.

FIG. 7 shows the format of character strings and variable values stored in working memory 309. Block 701 is a set of character strings, and represents that the preconditions or intermediate results represented by these character strings are satisfied.

Block 702 is a collection of variable value information, and (variable name><variable value>) indicates that the value of <variable name><variable value>.

Next, an overview of the proof processing 305 and derivation processing 306 will be explained. The proof process 305 is the control process 3
03, it is started with a character string representing the expected amount as an argument. First, if this predicted amount is stored in the working memory 309, the proof is successful. If it is not stored, a rule having this prediction amount as a conclusion character string is searched from the rule group section 308, and it is verified whether the condition character string and conditional expression of this rule hold true or not. Generally, multiple rules are found that have a predicted amount as a <conclusion string>, so these rules are verified one by one. If even one rule holds, the proof of foresight is a success; if all the rules do not hold, the proof of foresight is a failure.

Verification of the condition string> is performed by recursively repeating the same process as the verification of the expected quantity. That is, a rule having the condition character string as the conclusion character string is searched for, and whether this rule holds true or not is checked. If there is no rule with a conclusion string that is equal to this condition string, this condition string is displayed to the user and the condition expressed by this condition string is met or not. Inquire. To verify the conditional expression, the values of each variable included in the equality or inequality expressed by the conditional expression are stored in the working memory 309.
After finding this, calculate whether this equality or inequality holds true or not. If the value of the variable is not stored in the working memory 309, the variable name is displayed to the user and the user is asked about the value.

Note that details of the proof processing will be explained later using FIGS. 8 and 9.

The derivation process 306 refers to the character strings and variable values stored in the first working memory 309, finds a rule from the rule group part 308 that satisfies both the condition string and the conditional expression in the first condition part, and calculates this rule. This conclusion character string is extracted and compared with the solution candidate group in the solution candidate group 310 to determine whether it is a solution, and if it is a solution.

It is passed to the control processing 303, and if it is not a solution, it is stored in the working memory 309 as an intermediate result. Derivation process 306
In addition to the original input data, the system uses additional data newly input during the proof process, and also informs the user of the success or failure of the condition string and the values of variables as needed. match. The processing method of the derivation process 306 is similar to the conventional processing method.
+Addison-Wesley (1981) p p
239-250).

Next, details of the proof processing will be explained.

FIG. 8 shows a conclusion character string in which the proof process 305 exceeds expectations.
Represents the process of verifying whether a single rule holds true or not. The proof processing 305 is as follows.

This process is repeated until one rule is established or all rules have been verified. Note that the rules are intermediated as shown in Figure 5 and are assigned to the variable R (i.e., the variable ) is stored in the memory area allocated to R.

Block 801 assigns the condition part of the rule to variable CLi5t. As a result, a list of conditional character strings and conditional expressions is assigned to CLi5t, as shown in block 501 of FIG. These conditional character strings and conditional expressions are hereinafter referred to as conditions. Block 802
Assign the condition at the beginning of Li5t to variable C.

Block 803 determines whether C is a condition character string>such a conditional expression>. block 804 if the condition string is
~ block 809, block 81 if conditional expression>
By executing blocks 0 to 814, whether or not they hold true is verified.

Block 804 indicates that the condition string> is in the working memory 309.
Determine whether it is stored in the memory.

If it is stored, then the condition string is true and the process goes to block 816; if it is not stored, then
Go to block 805.

Block 805 verifies the <condition string> by recursively applying the proof process currently being explained to the rule having the <condition string> as the <conclusion string>. Block 806 determines whether the proof of C is successful or not. If successful, <condition string> is met and the process goes to block 816. If unsuccessful, block 807
go to.

Block 807 displays the <condition character string> to the user and inquires whether the fact represented by the character string holds true or not. If the user answers that the condition is true, the <condition character string> is true, and the process goes to block 809. If the user answers that the condition is not satisfied, the condition character string is not satisfied, and the process goes to block 815.

Block 809 is reached when the condition character string> is not in the working memory and cannot be proven, but it is determined that the condition is established based on an inquiry to the user.

In other words, this is a case where it is not possible to determine whether an event is established or not based on existing data, and it is determined that the event is established based on new data from the user. Therefore, in block 809, the condition character string> is stored in the working memory 309 so as to avoid repeating the same question to the user in the future. As a result, this <condition character string> can be determined to be established immediately by referring to the working memory 309 in the future. In particular, in the subsequent derivation process 306, when determining whether the condition character string is satisfied or not, it is only necessary to refer to the one working memory 309, and there is no need to ask the user again.

Blocks 810 to 814 show verification processing when condition C is a conditional expression intermediate word as shown in FIG. Block 810 assigns the first element of conditional expression intermediate word to variable Ex. Block 811 determines whether Ex is a get statement.

If it is a get statement, go to block 812; otherwise go to block 814.

Block 812 interprets and executes this get statement.

Find the variable value corresponding to <variable name> and assign it to <variable name>. This process will be explained in detail later using FIG. 9. Block 813 removes the executed get statement from C.

Block 814 is a conditional expression (<variable name 1>,
<Variable name 2>...<Variable name 1>) will be described below. By the time block 814 is reached, values have been obtained for all variable names in the conditional expression. Therefore, if the <conditional expression>(<variable name 1>, variable name 2>, variable name 1>) is satisfied,
It is possible to determine whether the application is unsuccessful. If true, go to block 816; otherwise go to block 8.
Go to 15.

Block 816 removes C1, that is, the condition found to hold, from CLi5t to the 0th order. Block 817 determines whether CLi5t is empty, that is, whether all conditions have been checked. If so, at block 818.

It is determined that the rule is established. If the search has not been completed, the process returns to block 802 to check the next condition.

Block 815 is processing when the condition is not met.

A rule is valid only if all conditions are met.

Therefore, here, it is determined that the rule does not hold.

FIG. 9 shows details of the get statement processing in block 812 of FIG. However, the get statement is in the form of get (<variable name>) as shown in FIG. 6, and is assigned to the variable Ex as described above. Block 901 is ge
Extract variable name〉 from the t statement and assign it to V Name. Block 902 determines whether the value of VName is stored in working memory 309. That is, the VName is stored in the working memory 309 shown in FIG.
It is determined whether or not the element Squirrel 1- is stored. If so, go to block 903.

If not, go to block 905.

Block 903 represents VNa+*e in working memory 309.
Assign a list whose first element is Item.

Block 904 assigns the second element of Item, ie, <variable value>, to VNan+e, ie, <variable name>.

Blocks 905 and 906 are processes performed when the value of variable name is not stored in the working memory 309. Block 905 displays the <variable name> to the user, inquires about its value, and assigns it to the <variable name>. Block 906 is
Similarly, in order to avoid having to ask the user again for the value of the variable name, the user should input the variable name and the variable value.
are made into a list and stored in the working memory 309. As a result, in the future, if you want to find the value of variable name> in the same way, you can refer to the working memory 309. In particular, after the derivation process 30
In step 6, when similarly obtaining the variable name, it is sufficient to refer to the working memory 309, and there is no need to inquire of the user again.

FIG. 1 shows the operation of the control process 303. Block 101 inputs variable names and values from the user prior to inference, forms a list of variable names and variable values, and stores them in working memory. 309. Block 102 is
Similarly, enter a character string and store it in the working memory 309.
The above variable values and character strings are data representing the situation or problem that the user, that is, the expert, wants to solve. Note that it is not necessary to input all necessary data in blocks 101 and 102; as described above, missing data can be input by the user during the certification process.

In block 103, the user inputs a predicted amount. The expected amount is a character string representing the user's judgment. block 1
04 starts the proof process 305 using the expected amount as an argument. The proof process 305 is performed by the processes shown in FIGS. 8 and 9.
It is verified whether the expected amount can be derived without contradiction from the data and the rules in the rule group 308. When the completion of the proof process and its results are reported, block 105 activates the derivation process 306. As described above, the derivation process 306 uses the data input in blocks 101 and 102 and the data input and stored in accordance with the timing of the proof process, and also uses the data input in blocks 101 and 102 and the data input and stored in accordance with the timing of the proof process. It derives its own solution while interrogating the data obtained and reports it to the control process.

Block 106 selects the output processing depending on whether the proof is successful or unsuccessful. If the proof is successful, block 10
Block 7 outputs a message expressing approval, and then block 108 compares the solution received from the derivation process 306 with the expected amount, and if different, outputs it as a different answer. If no solution different from the expected amount is obtained, a message to that effect is output.

On the other hand, if the proof fails, block 109 outputs a message expressing the contrary, and then block 11
0 outputs all solutions received from the derivation process 306, if any, as rival solutions.

If no competing solution is obtained, a message to that effect is output. Messages for approval or opposition as well as different or counter solutions are obtained through proof processing 305 and derivation processing 30.
You can output it every time you receive a report from 6, or you can output it every time you receive a report from memory 3.
07, and after the image processing is completed, they are combined to assemble an appropriate message group.
However, in general, the derivation process takes a considerable amount of time, so considering the psychology of the user,
It is preferable to output it every time a report is made.

Next, using a specific example of medical diagnosis, we will show how an inference machine verifies and supplements the judgment of an expert using the above processing method. It is assumed that the inference machine has already input a set of rules for medical diagnosis, converts this into intermediate language, and stores it.

The user is a doctor, who uses the patient's symptoms as data and inputs his or her own diagnosis result, that is, the name of the disease, into the inference machine as a prediction amount. The inference machine attempts to prove this conjecture from the data using the rules, and expresses approval if the proof is successful, and expresses opposition if it fails. Furthermore,
A solution, that is, a disease name is derived from the data, and this is output as a different solution or a counter solution.

In the first example, the user, ie, the doctor, determines that the patient has influenza based on the symptoms of ``feeling tired and has a body temperature of 38.5 degrees Celsius,'' and consults the reasoning machine regarding the determination.

FIG. 1O is an example of a rule set used in this example problem. Blocks 1001 to 1004 each represent one rule. Block 1001 indicates that if the user feels tired and his/her body weight is 2 kg or more less than normal, the user is in a state of exhaustion. ``Feeling tired'' is a conditional string.

'#Current weight' and '#Normal weight' are variable names,
Current weight-#normal weight<2.0' is a conditional expression regarding these variables. In this example, # will be added to the beginning of the variable name.

Block 1002 is in an exhausted state and has a body temperature of 38.0.
If it's higher than ℃, it indicates influenza. Block 1003 indicates pancreatitis if the patient is exhausted, has abdominal pain, and has a high temperature above 37.0°C. Block 1004 indicates tuberculosis if the patient is exhausted, the temperature is higher than 37.0°C and lower than 38.0°C, and the duration of symptoms is longer than 30 days.

FIG. 11 is an intermediate word resulting from analysis and conversion of the rules in FIG. 10 by the analysis and conversion process 304 in FIG. Blocks 1101 to 11o4 are blocks 1001 to 1, respectively.
This is a rule intermediate word for 004. These intermediate words are stored in the rule group section 308 in FIG.

Further, in the solution candidate group section 310, a set of disease names that can be solutions for medical diagnosis under the rules of FIG. 10 is stored in advance. 'Influenza', 'pancreatitis' and 'tuberculosis' are remembered here.

When the user starts the inference machine, the control process 303 in FIG. 3 starts the process in FIG. 1. block 1
01 inputs variable values from the user and stores them in the working memory 309. Here, #body temperature and its value 38.5
Enter and save.

Next, block 102 inputs a character string from the user,
It is stored in the working memory 309. Here, ``feeling of fatigue'' is input and memorized. The contents of the working memory 309 at the end of the processing of block 102 are shown in FIG.

Block 103 is the predicted group, 'influenza'
Enter. Block 104 initiates proof processing 305.

The operation of the proof processing 305 will be described below with reference to FIGS. 8 and 9.

In the proof processing 305, first, when checking whether the predicted solution 'influenza' is stored in the working memory 309, 'influenza' is not stored in the working memory 309. Next, 'influenza' is determined as a conclusion. The rule intermediate word to be converted into a character string is searched for in the rule group section 308. In this example, block 1102 in FIG.
Process the diagram.

Block 801 takes out the conditional part of R and CLi5t
Assign to . Here, block 1106 is added to CLi5t.
is assigned. Block 802 assigns to C the leading element of CLi5t, ie, the 'exhausted state'. Block 8
03 determines whether C is a conditional string or a conditional expression. Here, it is determined that it is a conditional string.

Block 804 determines whether C or 'exhausted condition' is stored in working memory 309. Here, since the working memory 309 is in the state shown in FIG.
becomes.

Block 805 then attempts to prove the 'exhausted state'. That is, the proof process 305 is started recursively using the 'exhaustion state' as the expected quantity. below.

The operation of the proving process 305 for proving the 'exhausted state' will be explained.

The proof process 305 searches for a rule whose conclusion character string is ``exhaustion state''. Here we find block 1101. Therefore, the process shown in FIG. 8 verifies whether the rule holds true or not. Block 801 assigns block 1105 to CLi5t.

Book 802 assigns ``feeling of fatigue'' to C.

Block 803 determines that C is a condition string.

Block 804 determines whether C, ``Feeling Fatigue'', is stored in working memory 309. Here, since the working memory 309 is in the state shown in FIG.
It becomes es. Block 816 removes the processed condition, ie, 'feeling tired', from CLi5t. As a result, CLi5t contains (get(#actual body type)get(#normal weight)#current weight-existing normal weight<2. 0)) is left. This is.

1 consisting only of (get(#actual type)...<2.0)
It is a list of elements.

Block 817 determines whether all conditions have been determined. Here, the answer is No.

Block 802 assigns the first element of CLi5t to C. Here, in C, (get (#actual type) get
(#normal weight) #current body type - #normal weight<2.0) is substituted. Block 803 determines that C is a conditional expression.

Block 810 adds the first element of C, ie, get(
#Substitute actual type). Block 811 is Ex is ge
It is determined that it is a t sentence. Block 812 executes Ex. The execution of this get statement will be explained below using FIG. 9.

Block 901 sets the argument of Ex, that is, the #actual type to V
Assign to Nams. Block 902 determines whether the value of VName is stored in working memory 309. Here, the determination is No.

Block 905 queries the user for the value of VName or #actual type. Here, assume that the user answers 55. Therefore, 55 is assigned to VName. Block 906 stores the value of VName in working memory 309. Here, a list called (#actual type 55) is stored.

This completes the processing of the get statement and returns to block 813 in FIG. Block 813 removes the processed get statement from C. As a result, C has (get (
#Normal weight) #Current body type - #Normal weight <2.0) is left. Block 810 gets to Ex (#normal weight)
Substitute. Block 811 determines that Ex is a get statement.

Block 812 performs the same process as described above.

Execute this get statement. # Since the normal weight value is not stored in the working memory, ask the user again.

Here, it is assumed that the user inputs 60. Therefore,#
60 is substituted for the normal weight, and the working memory 309 contains (
#Normal weight 60) is stored.

Block 813 removes Ex from C, leaving C with (#actual body type - #normal weight<2.0).

Block 810 adds #actual type to #normal weight2.
Substitute 0. Since this is not a get statement, block 811 determines No. Block 814 uses the values of #current body type and #normal weight determined so far to
Establishment of Ex.

Determine failure. Here, it is determined that it is true.

Block 816 removes the processed conditions from CLi5t. CLi5t is (get(#actual type)・(2
, 0), the result of processing one line is an empty list. Block 817 determines whether all conditions have been processed.

Here, the answer is Yes.

Block 818 determines that the rule, block 1101, holds true. As a result of the above, the 'exhausted state' is successfully proven, and block 8 in the process of block 1102
Return to 06.

After block 806 determines that the proof of C, ie, the 'exhausted state', is successful in the above manner, block 816 removes the processed condition, ie, the 'exhausted state', from CLi5t.As a result, CLi5t is ((get
(#body temperature) #body temperature>38.0)) is left. Block 817 determines whether all conditions have been examined. Here, the answer is No.

Block 802 returns C to (get(#body temperature)#body temperature>3
8.0). Block 803.

It is determined that C is a conditional expression. Block 810 is
Assign get(#body temperature) to . Block 811 is E
It is determined that x is a get statement. Block 812 executes this get statement by processing similar to that described above. At this time, since (#body temperature 38.5) is stored in the working memory 309, no inquiry is made to the user. As a result of this processing, 38.5 is assigned to #body temperature.

Block 813 assigns (#body temperature>38.0) to C. Block 810 indicates Ex #body temperature>38.0
Substitute. Block 811 determines NO.

Block 814 indicates that Ex or #body temperature>38.0
It is determined that the following holds true. Block 816 is CLi5t
Remove the condition that has been processed from (, CLi1it is (
get(#body temperature) #body temperature>38.0), CLi5t becomes empty as a result of this process. Block 817 determines that all conditions have been processed. Therefore, block 818 determines that the rule being processed, ie, block 1102, holds true.

As a result of the above, the proof process 305 succeeds in proving the predicted solution 'influenza', and the control process 303
Report to. The contents of the working memory 309 at this point are:
It is shown in FIG.

Here, the block 105 in FIG. 1 is the derivation process 306
Start. The derivation process 306 uses the rule set in the rule group section 308 to derive a solution from the data in the working memory according to a conventional processing method, and outputs this to the control process 303.

If there are multiple possible solutions, all are output.

First, block 1101 in FIG. 11 is established.

Determine failure. In the derivation process, a rule in which all the conditions in the condition part are satisfied is determined to be true, and a rule in which even one condition is not satisfied is determined to be untrue.

Since all conditions are met for block 1101, it is determined that the rule is met. In this determination, the values of #actual body type and #normal weight are required, but the proof process 30
These values input by the user in step 5 are stored in the working memory 309, so there is no need to inquire about them again. As a result of block 1101 being established, the fact that there is a "depleted state" is established. Here, by referring to the solution candidate set in the solution candidate group 310, it is determined whether the 'exhausted state' can be a solution. Here, the answer is NO.

Therefore, using the "exhausted state" as an intermediate result, the working memory 3
Stored in 09.

Next, it is determined whether the block 1102 is established or not, and a result that the block 1102 is established is obtained. As a result, the fact of ``influenza'' is established. Referring to the solution candidate group 310, it is determined that 'influenza' can be a solution, and the result is passed to the control process 303.

Next, it is determined whether block 1103 is satisfied or not. Here, the user is asked whether "I have abdominal pain" is true or not. Suppose that the user answers that it is true. in this case,
Block 1103 is established, and the fact of "pancreatitis" is established. Referring to the solution candidate group 310, it is determined that ``pancreatitis'' can be a solution, and this is determined in the control process 303.
Output to.

Finally, it is determined whether block 1104 holds true or not.

Here, since #body temperature is 38.5, it is not established.

During this time or after the completion of the derivation process, the control process 30
3, if the report from the proof processing 305 is 1 success,
Through the processing in blocks 106 and 107, a message indicating that the user agrees with the decision is output. block 108
compares each interval obtained in the derivation process 306 with the expected amount, and since 'influenza' is the same as the expected amount, it is discarded, and 'pancreatitis' is different from the expected amount, so pancreas is treated as another diagnosis. Outputs a message indicating that there is a possibility of fire.

Next, a case will be described in which the inference machine disagrees with the user's judgment and outputs a counter-judgment. This time, the user
Based on the symptoms of ``feeling tired'' and #body temperature of 37.5 degrees Celsius, it is determined that it is influenza, and the inference machine is consulted regarding the determination. It is assumed that the rule group section 308 stores the rule set shown in FIG. 11, as in the example above.

As in the previous example, control processing 303 starts the processing of FIG. block 101 and block 10
Through the process 2, the contents of the working memory 309 become as shown in FIG. Block 103 inputs the expected amount 1 influenza.

Block 104 initiates proof processing 305.

In the proof process, a rule having 'influenza' as a conclusion character string, that is, block 1102, is found, and the process shown in FIG. 8 verifies whether the rule holds true or not. block 80
1 assigns block 1106 to CLi5t. Block 802 assigns ``exhaustion state'' to C. Block 803 determines that C is a condition string. Block 804 determines that the 'exhausted state' is not in working memory 309.

Block 805 recursively activates proof processing 305 to attempt to prove "exhausted state". Proof processing 305 establishes block 1101 with "exhausted state" as the conclusion character string by processing similar to the above. .

Verify failure. Here, as in the previous example, consider the case where block 1101 is established. In this certification process, the value 55 of #actual body type and the value 60 of #normal weight are input by the user and stored in the working memory 309. The contents of the working memory at this point are shown in FIG.

1 exhaustion state' is successfully proven and returns to block 806 in the processing of block 1102, block 806
Determine Yes. Block 816 causes CLi5t
, ((get(#body temperature) #body temperature>38.0)) is left. Block 817 determines NO.

Block 802 returns C to (get(#'body temperature)#body temperature>
38.0 ). Block 803 determines that C is a conditional expression. Block 810 assigns get(#body temperature) to Ex. Block 811 determines Yes.

Block 812 executes get(#body temperature) and assigns 37.5 to #body temperature. Block 813 leaves C with (#body temperature>38.0).

Block 81o assigns E x (temperature) 38.0. Block 811 determines No. Block 814 determines that Ex does not hold. As a result, it is determined that block 815 is R, that is, block 1102 is not established.

Since there is no rule other than block 1102 that uses "influenza" as the conclusion string, the proof processing 305 fails to prove "influenza" and reports this to the control processing 303.

Block 105 then initiates a derivation process 306. In the derivation process 306, first, the establishment of the block 1101,
Verify failure. In this example, it is determined that it holds,
The 'exhaustion state' is stored in the working memory 309. Next, it is verified whether block 1102 holds true or not. ＃Body temperature is 3
Since it is 7.5, it is determined that it does not hold true.

Next, it is determined whether block 1103 is satisfied or not. Here, the user is asked whether ``I have abdominal pain'' is true or not. Suppose that the user answers that it is true. in this case,
Block 11o3 is established, and the fact of "pancreatitis" is established. Referring to the solution candidate group 310, it is determined that 'pancreatitis' can be a solution, and this is output to the control processing 303.

Next, it is verified whether block 1104 holds true or not. Here, ask the user how many days #symptoms last. Assume that the user answers 50. In this case, block 11
04 is established. Referring to the solution candidate group 310, it is determined that "tuberculosis" can be a solution, and "tuberculosis" is output to the control processing 303.

During this time or after the completion of the derivation process, the control process 30
3, if the report from the proof processing 305 is “Failure 1,”
Through the processing in blocks 106 and 109, a message to the effect that the user disagrees with the decision is output. block 110
adopts all the solutions obtained by the derivation process 306 and outputs a message indicating that there is a possibility of pancreatitis or tuberculosis.

As shown in the above example, this embodiment verifies the expert's judgment, and at the same time, if the expert agrees with the judgment,
In addition, if the judgment is contrary to the judgment, a counter solution, if any, is automatically presented, thereby preventing experts from making premature decisions or making oversights, and leading to a more valid conclusion more efficiently.

Next, enter multiple predicted solutions that represent the expert's judgment,
An example will be explained in which each of these predicted solutions is verified, and depending on the result, approval or opposition to each predicted solution is expressed, and a solution different from any of the predicted solutions is output as a different or counter solution. .

The configuration and processing operation of this embodiment are as follows:
The components other than 03 are the same as those of the first embodiment described in FIGS. 1 to 9.

FIG. 16 shows the operation of the control process in this embodiment, and is an alternative to FIG. 1. block 1
Blocks 01 and 102 are the same as blocks 101 and 102 in FIG. 1, and input data from the user.

Block 1601 reads a plurality of child sizes as a list and assigns them to SLi5t. Block 1602 is
Take out the first predicted amount from S Li5t and substitute it into S. Block 1603 is a proof process 30 using S as an argument.
Start 5. The operation of the proof processing 305 is similar to that of the first embodiment.

Block 1604 removes the processed prediction amount S from 5List. Block 1605 determines whether all prediction quantities have been processed.

If the processing has been completed, the process goes to block 105; if the process has not been completed, the process returns to block 1602 to process the next expected amount.

Block 105 initiates the derivation process. This derivation process is the same as the derivation process 30G in the first embodiment.

After the derivation process is completed or in parallel with it, the output process is performed. Block 1606 determines whether the proof of at least one conjecture was successful or the proof of all conjectures failed. In the former case, block 1607 is
A message expressing approval or disapproval for each predicted amount is output, and block 1608 removes all the word spacings that are equal to the predicted amount from the word spacing received from the derivation process 306 and outputs the rest as separate spaces. In the latter case, block 1609.

Outputting a message of opposition to all predicted solutions, block 1610 outputs all solutions received from derivation process 306 as counter solutions.

As a variation, each time the processing result of one conjecture quantity is received from the proving process 305 (eg, before block 1604), the approval or disapproval of that conjecture quantity may be output. In this case, the decision in block 1606 is made after all predicted solutions have been processed, and the processing in block 1608 or 1610 is performed according to the result.

As described above, according to this embodiment, a plurality of prediction quantities representing the expert's judgment are input, each of these prediction quantities is verified, and depending on the result, the judgment is made for or against each prediction quantity. A solution that differs from any predicted quantity can then be output as an alternative or counter solution.

Therefore, even if an expert has come up with multiple decisions or is unsure about one decision, the inference machine can verify each decision by inputting each of the possible decisions as prediction quantities.

By outputting an alternative or counter solution that is different from any judgment, it is possible to compensate for an expert's oversight and efficiently lead to the most appropriate conclusion.

Next, an embodiment will be described in which instead of directly outputting a different answer or a counter solution to the expert's judgment, information indicating the existence of a different answer or a counter solution is output.

The hardware configuration of this embodiment is as shown in FIG. 2, and is the same as that of the first embodiment.

FIG. 17 shows the functional configuration of this embodiment. block 30
10 is a processing group, including input/output processing 3o2° control processing 3030. Analysis conversion processing 304゜Proof processing 305.
and derivation processing 3060.

The block 3070 is a memory, and the rule section 308,
Working memory 309. Solution candidate group 310 and solution memory 1
Consists of 701.

Among these, input/output processing 302. Analysis conversion process 304
．． Proof processing 305. Ruhr County Department 308゜Working Memory 30
9 and the solution candidate group section 310 are the same as those in the first embodiment.

The control processing 3030 receives data from the input/output processing 302 and stores it in one working memory 309, as in the first embodiment. When receiving the predicted amount and the activation command from the input/output processing 302, the proof processing 305 is started using the predicted amount as an argument, and depending on the result of the proof, an expression for or against the predicted amount is sent to the input/output processing 302. Output. Next, derivation processing 3060 is activated. Derivation process 3
After completing step 060, it is determined whether the solution memory 1701 stores a solution different from the expected amount. If stored, the existence of a different or counter solution is indicated to the user via the input/output processing 302, and furthermore, the solution in the solution memory 1701 is output to the user in response to a request from the user. .

When the derivation processing 3060 obtains a solution, instead of outputting it to the control processing 3030, the derivation processing 3060 stores the solution in the solution memory 170.
1, and when the processing is completed, it is reported to the control processing 3030. Except for this point, the process is the same as the derivation process 306 in the first embodiment.

Solution memory 1701 receives the solution from derivation process 3060 and stores it.

FIG. 18 shows the operation of the control processing 3030 of this embodiment. Blocks 101 to 107 and 109 are blocks 101 to 10 in FIG. 1 in the first embodiment.
7 and 109, the user inputs data and predicted solutions, verifies them in proof processing 305, and then starts derivation processing 306. Then, based on the verification results, express approval or disapproval of the predicted solution. As mentioned above, the derivation process 307 stores the obtained solution in the solution memory 17.
The process is stored in the previous file and the end of the process is reported to the control process 3030.

If the proof of the predicted solution is successful, block 1801 checks whether there is another solution in the solution memory 17o1, that is, a solution different from the predicted solution, and if not, block 180
2 outputs a message to the effect that the Betsuma does not exist, and the process ends.6 On the other hand, if the Betsuma is in the solution memory 1701, the block 1803 outputs a message to the effect that the Betsuma exists, and the block 18o4 asks the user if he would like to request another output. If output is not requested, the process ends, and if output is requested, block 1805 reads the contents of the solution memory 1701 and outputs only those that differ from the expected solution as separate items. Processing ends.

If proof of the predicted solution fails, block 1806 checks whether a solution is stored in solution memory 1701. In this case, there is no need to compare with the expected solution; if there is a solution, it is a counter solution. If there is no competing solution, block 1807 outputs a message to that effect and the process ends.

If there is a counter solution, block 1808 outputs a message to that effect, and block 1809 asks the user whether or not the counter solution should be output. If output is not requested, the process ends; if output is requested, block 1810 outputs the contents of solution memory 1701 as a counter solution, and the process ends.

As described above, the present embodiment not only verifies the user's judgment, but also indicates the existence of a different answer or counter solution to the user's judgment, and can output the different answer or counter solution according to the user's wishes. The experts are.

Depending on your work situation, you can look into separate or counter solutions, or you can ignore them.

In the embodiments described above, the derivation process is started after the proof process is completed. However, both of these treatments
May be done in reverse order. Furthermore, they may be executed in parallel. For example, alternating time slots may be assigned to both processes, allowing periodic time-sharing operations to occur under the control of a timer. Alternatively, priority may be given to one of the processes, and the other process may be executed only while this process is in a waiting state for, for example, making an inquiry to the user, and only after the process has finished. In such parallel processing, information acquired and stored in the working memory by either process, either as an intermediate result or by a request to a user, can be used by the other process. Generally, it is preferable to give priority to certification processing. The reason for this is that the result of the proof process affects the handling of the solution obtained by the derivation process, and the sooner the user is notified of the validity of the predicted solution, the less the user will be irritated.

FIG. 19 shows an example of a hardware configuration particularly suitable for parallel processing of proof and derivation. The input/output terminal 201 and the storage device rI 1203 are equivalent to the input/output terminal 201 and the storage device 203 in FIG. 2, respectively. The control unit 1901 performs input/output processing 3o2. This is a unit that executes control processing 303 (or 3030 in FIG. 17) and analysis conversion processing 304, and upon receiving a startup command, simultaneously activates proof processing unit 1902 and derivation processing unit 1903. Proof processing unit 19
02 is dedicated hardware for executing the proof processing 305 in FIG. 3, and the derivation processing unit 1903 is
This is dedicated hardware for executing the derivation process 306 in FIG. 3 (or 3060 in FIG. 17). The storage control device 1904 includes control units 1901. Coordinates storage access request conflicts between the proof processing unit 19o2 and the derivation processing unit 1903. In order to speed up processing, a cache memory may be provided in the storage control device 1904, and either the proof processing unit 1902 or the derivation processing unit 1903 (preferably the proof processing unit) is a control unit. It may be configured as part of 19o1.

Furthermore, in the embodiments described above, the control process 303 or 3030 compares the solution obtained by the derivation process with the expected solution. However, the predicted solution may be passed to the derivation process 303 or 3030 and the derivation process itself may perform this comparison.

Next, a detailed explanation of the present invention to an expert system will be given. An expert system is realized by equipping an inference machine with a set of rules in advance. Therefore, the aforementioned embodiments can be used as they are because the set of rules is stored in advance and the process of inputting the set of rules is removed.

A corresponding example will be given in an expert system.

Thus, the invention can provide a competent assistant system, either by a reasoning machine or an expert system.

〔Effect of the invention〕

As is clear from the above description, according to the present invention, the verification of the user's judgment and the presentation of alternative or counter solutions are automatically performed with a single activation operation, thereby preventing premature Overlooks are prevented and the most reasonable conclusion can be reached efficiently. The proof process and the derivation process share data, thus preventing duplication of queries to users for additional data. Efficiency can be further improved by performing the derivation process while waiting for the proof process.

A variant that processes multiple predicted solutions at once is useful when the user is confused between multiple decisions, and a variant that first outputs only the presence or absence of a separate or counter solution may overwhelm the user with excessive information. save from

Further, according to the apparatus of the present invention, complete simultaneous execution of proof processing and derivation processing is realized.

Processing time is reduced.

[Brief explanation of the drawing]

FIG. 1 is a flowchart showing the procedure of control processing in an embodiment of the method of the present invention, and FIG. 2 is a block diagram showing the hardware configuration of an example of a data processing system suitable for executing the method of the present invention. FIG. 3 is a block diagram showing the functional configuration of a data processing system that executes the embodiment, FIG. 4 is a diagram showing an example of the format of rules used in the present invention, and FIG. The figures are diagrams showing the format of the intermediate word of the rule shown in FIG. 4, FIG. 6 is a diagram showing details of the conditional expression in the rule intermediate word shown in FIG. 5, and FIG. 8 is a diagram showing an example of the format of the contents of the working memory used in the present invention, FIG.
10 is a flowchart showing a procedure for determining variable values in the processing procedure shown in the figure, FIG. 10 is a diagram showing an example of a rule set, and FIG. 11 is a diagram showing an intermediate word of the rule set in FIG. FIG. 12 is a diagram showing an example of the contents of the working memory at the beginning of the inference process, and FIG. 13 is a diagram showing the contents of the working memory that have changed from FIG. 12 at the middle of the inference process. Figure 14 is a diagram showing another example of the contents of working memory at the initial stage of the inference process, and Figure 15
The figure is a diagram showing the contents of the working memory changed from FIG. 14 in the middle of the inference process, and FIG. 16 is a flowchart showing the procedure of control processing in the second embodiment of the method of the present invention. FIG. 17 is a block diagram showing the functional configuration of a data processing system that implements the third embodiment of the present invention. FIG. 18 is a flowchart showing the procedure of control processing in the third embodiment, and FIG. 19 is a block diagram of an embodiment of the judgment assisting device of the present invention suitable for executing the method of the present invention. 101-103... Input of data and predicted solutions, 104.
...Start of proof processing, 105...Start of derivation processing. 107... Output to the effect that I agree with the predicted solution, 108... Output to the effect of another solution, 109... Output to the effect that I disagree with the predicted group, 11
0... Output of counter solution, 1601-1605... Input of multiple child calculations, 1801-1803... Output of presence/absence of the difference, 1804.1805... Output of the difference according to request, 1806- 1808... Output of presence/absence of counter solution, 1
809.1810...Output of counter solution according to request, 19
01... Control unit, 1902... First inference unit for proof processing, 1903... Second inference unit for derivation processing, 201... Input/output unit 1-1
203...Shared memory.

Claims (1)

[Scope of Claims] 1. Comprising an input device, an output device, a storage device including a rule storage unit that holds one or more rules, and a processing device including an inference means for performing inference processing based on the rule. In the data processing system, an input step of receiving data representing a problem and a predicted solution to the problem via the input means and storing the data in the storage device; and applying the rule to the data by the reasoning means to generate the prediction. a proving step for attempting to prove a solution; a deriving step, which is started in conjunction with the proving step, and in which the inference means applies the rules to the data to derive all possible solutions to the problem; a first output step of outputting information indicating success or failure through the output device; and outputting at least a part of the derived solution through the output device in association with the success or failure of the proof. Second
and an output step, and at least one of the proof step and the derivation step includes a step of requesting additional data that becomes necessary in the process via the output device, and receiving the additional data via the input device. and storing the additional data in the storage device, and the other of the proving step and the deriving step uses the additional data stored in the one process as necessary. 2. In claim 1, the second output step outputs only the derived solutions that are different from the expected solution as different solutions when the proof is successful, and when the proof is unsuccessful. A decision aiding method that outputs all of the derived solutions as competing solutions. 3. The decision assisting method according to claim 1 or 2, wherein the input step receives a plurality of predicted solutions, and the proving step attempts to prove each of the plurality of predicted solutions. 4. In claim 1, in place of the second output step, outputting information indicating the presence or absence of a solution different from the predicted solution, and outputting the solution different from the predicted solution in response to a request via the input device. A decision aiding method comprising the step of outputting via an output device. 5. In claim 1, 2, 3, or 4, the proving step includes a step of requesting additional data that becomes necessary in the process via the output device, and receiving the additional data via the input device. and storing the additional data in the storage device, and the deriving step is started after the proving step is completed, and the stored additional data is used as necessary. 6. In claim 1, 2, 3 or 4, the proving step and the deriving step are performed in parallel, and each of both steps requests additional data via the output device as required during the process. and receiving the additional data via the input device and storing it in the storage device, and further using the additional data stored by the other device as needed. 7. The decision aiding method according to claim 6, wherein the deriving step is performed while the proving step is in a waiting state. 8. Any one of claims 1 to 7, further comprising the step of receiving at least a portion of the rules via the input device and storing them in the storage device, thereby forming the rule storage. Decision aid method. 9. an input unit, an output unit, a first inference unit that attempts to prove a predicted solution by applying a set of rules to data representing a problem; and a first inference unit that attempts to prove a predicted solution by applying a set of rules to data representing the problem; a second inference unit that derives a solution; and a control unit that is connected to the input unit, the output unit, and the first and second inference units to control parallel operations of the first and second inference units and performs input/output processing. and a shared memory connected to the first and second inference units and the control unit to store necessary information including the data and the rule group.